Deep learning method improves privacy in image and speech transmission
Privacy-Preserving Deep Joint Source-Channel Coding with In-Loop Concept Erasure
Machine Learning
Summary
Sending pictures or speech through communication channels can accidentally reveal sensitive information like gender or race. The authors developed a technique called LEAPSC that removes such private details during transmission while keeping the main task accurate. They tested it on popular datasets of faces and speech and showed it reduced privacy leaks better than existing methods. Their approach makes it harder for attackers to guess private traits from transmitted data.
What this means in practice
- •For mobile communication engineers: Protect sensitive user attributes while transmitting images or audio over wireless channels with minimal loss in task performance.
- •For speech recognition developers: Improve privacy in transmitted voice commands by reducing speaker identity leakage during feature encoding.
Authors
Rami Eid, Maria Slim, Mariette Awad, Hadi Sarieddeen
Abstract
Deep joint source-channel coding (DeepJSCC) transmits learned semantic features efficiently but can leak sensitive attributes such as gender, race, or speaker identity. We propose LEAPSC (LEACE-in-the-loop privacy for semantic communication), whose core contribution is the integration of in-loop least-squares concept erasure (LEACE) within a variational information bottleneck (VIB) encoder. By periodically refitting the projection operator during training, LEAPSC couples the encoder dynamics to the erasure mechanism, driving attribute-conditional mean differences toward zero within each task-label group on the fitting sample. Additional components, namely conditional value-at-risk (CVaR) tail-sensitive privacy, feature-wise linear modulation (FiLM) signal-to-noise ratio conditioning, and Lagrangian dual ascent, improve robustness across channel conditions and over the high-leakage tail of samples. On CelebA, FairFace, and Google Speech Commands, LEAPSC reaches task accuracy of 0.862, 0.755, and 0.925 respectively, with attacker accuracy at or below the label-only floor on CelebA (0.548 vs. floor 0.580) and within 2 percentage points (pp) of chance elsewhere, improving over an information-bottleneck adversarial baseline (IBAL) at a matched 52-epoch budget by +3.6, +2.5, and +1.3 pp (Welch's t-test, p=0.019 on CelebA).